Re: Can't diff after killing svnlook

[me, whining about how long "svnadmin recover" took]
> Actually, "strace -p <pid>" works on Linux to see if something is
> hanging in select. Luckily it wasn't: "svnadmin recover" *finally*
> completed after almost two hours. That's a lot of repository downtime
> to suffer just because I hit Ctrl-C at an inopportune moment...

[Greg Stein commiserates]
> Yah, it is :-(
>
> I haven't seen one take that long before. I wonder how much of it was spent
> doing work versus how much time was "hmm. there is a lock. let's wait to see
> if it gets used. nope. clear it. oh, another lock..."

I did attach an strace to the svnadmin recover at a few points. It
looked like most of what it was doing, from a system call PoV, was
opening the data files in the repository/db directory ('nodes',
'strings', etc.), doing stuff on them (I recall lots of fcntl()s),
and closing them again and again and again and again and ...

(That's from memory of three days ago, BTW.)

> In any case, I could see an interesting doc item would be some "expected"
> times for running recovery based on the DB size. I wonder if there is
> correlation with static items like numbers of items, and the correlation to
> runtime items like how many locks were left open, etc.

Would also be nice for "svnadmin recover" to print some progress info as
it's going. Obviously, "X% done" would be best, but just some
indication of activity would do.

        Greg

-- 
Greg Ward - software developer                gward@mems-exchange.org
MEMS Exchange                            http://www.mems-exchange.org
---------------------------------------------------------------------
To unsubscribe, e-mail: dev-unsubscribe@subversion.tigris.org
For additional commands, e-mail: dev-help@subversion.tigris.org